Re: Knowledge Graphs and Embedding Vectors: A New MIME Type for Numeric Data Tensors?

Sebastián,

Thank you. I will take a look at that document.

I revised the preliminary example #4 (https://github.com/AdamSobieski/Web/discussions/5) and, now, the two envisioned new MIME types are:

application/tensor+json
application/tensor+cbor

MIME-type parameters for tensors' types and shapes aren't required, after all, in my opinion, as these can be properties and values in the JSON and CBOR data...


Best regards,
Adam

________________________________
From: Sebastian Samaruga <ssamarug@gmail.com>
Sent: Friday, August 14, 2026 1:28 AM
To: Adam Sobieski <adamsobieski@hotmail.com>
Cc: W3C Semantic Web IG <semantic-web@w3.org>
Subject: Re: Knowledge Graphs and Embedding Vectors: A New MIME Type for Numeric Data Tensors?

Don't know if this is of any use. Regards,

Sebastián.


On Fri, Aug 14, 2026, 2:06 AM Adam Sobieski <adamsobieski@hotmail.com<mailto:adamsobieski@hotmail.com>> wrote:
Semantic Web Interest Group,

Hello. I am excited to share some ideas for storing embedding vectors alongside knowledge graphs referring to them.

Earlier, in this mailing list, we discussed multiple approaches for representing numeric data tensors within RDF, including how to go about doing so in RDF literals.

Embedding vectors can also be viewed as resources; embedding vectors, then, can be stored in message parts within multipart MIME messages.

For these and other purposes, two new MIME types are envisioned:

text/numeric-data-tensor
application/numeric-data-tensor

Drawing inspiration from the RDF Tensor project (https://neverblink-oss.github.io/rdf-tensor/), one could specify tensors' types and shapes using MIME type parameters.

text/numeric-data-tensor; data="float32"; shape="[1024]"
application/numeric-data-tensor; data="float32"; shape="[1024]"

With MIME types for numeric data tensors, one could store these resources in multipart MIME messages and utilize the "cid:" URL scheme (RFC 2392) from within knowledge graphs in these messages to refer to the other message parts.

Please feel free to check out a preliminary example, example #4, at: https://github.com/AdamSobieski/Web/discussions/5 .

Hopefully interesting. What do you think about these ideas? Thank you.


Best regards,
Adam Sobieski